When it comes to AI shopping assistants, experts say data readiness and security — not speed — will separate winners from cautionary tales.
Consumer adoption of AI shopping assistants accelerated during the holiday season, with 1 in 5 purchases driven by AI-powered tools and assistants, influencing a whopping $262 billion in revenue, according to Salesforce.
Brands that launched their own shopper agents before the holiday saw significantly more growth than those that hadn’t, said Caila Schwartz, director of consumer strategy and insights at Salesforce.
That momentum is putting pressure on CX leaders to act. But experts say the biggest risk isn’t falling behind — it’s deploying before the organization is ready.
AI shopping assistants promise to improve the shopping experience by delivering personalized, conversational and efficient customer experiences.
Before deploying, however, brands need to get their data strategy right. Unified data is the most critical prerequisite, experts say. Agents need full context on both the customer and the product catalog, and that data is rarely in one place. Product information, for instance, may live in a product information management system, while an enterprise resource planning system keeps track of inventory, and user manuals contain detailed product information.
“Your outcomes are really as good as the inputs that you provide,” said Grant Deken, head of product at Klaviyo, an AI marketing and CRM provider. Deken compared training an AI agent to onboarding a new employee.
Brands also need to invest in data hygiene, content quality and the right integrations before expecting strong results. Deploying before unifying customer, product and brand data risks undermining the personalization the technology is supposed to deliver.
“If the assistant recommends out-of-stock items, misstates policies or misrepresents compatibility, trust erodes quickly,” Instacart CTO Anirban Kundu said in an email. But there’s “a cold start challenge — limited early data can hurt performance when first impressions matter most.”
In addition, while it’s possible to develop an AI shopping assistant in-house, most brands don’t have the resources to do it well. Some businesses attempted internal builds before coming to Zoovu, an AI-powered product discovery platform,xf said, describing their prior efforts as little more than science experiments.
Security is another key reason to rely on vendors, with shopping assistants facing constant prompt-injection attacks as people try to get them to offer free products or leak sensitive data.
“You can’t protect yourself from prompt injection with clever prompts,” Taylor said.
AI agents that fit into the overall experience
CX leaders must carefully evaluate vendors to ensure they can deliver on their promises. In addition to demonstrating that their technology drives measurable sales outcomes, vendors should also be able to show how their technology fits into the entire customer lifecycle rather than treating an agent as an isolated tool. The signals generated by shopping interactions, including from customers who don’t buy, can feed back into marketing, segmentation and future engagement.
“That’s not something a lot of people are thinking about or talking about,” Deken said.
So, rather than a full-scale rollout, Taylor recommended starting with a single product category to learn how customers engage with it.
“Start in a very scoped, clearly measurable project, and then build a business case from that,” Taylor said.
Experts also warned against a set-it-and-forget-it mindset, noting that brands need to regularly refresh their knowledge bases and add new capabilities over time.
“Building a reliable assistant isn’t a feature launch, but an ongoing engineering commitment across [machine learning], infrastructure and product,” Kundu said. “Maintenance never stops: catalogs change, promotions rotate, rules evolve and foundation models update in ways that shift behavior. Continuous testing and tuning become permanent disciplines.”
Brands must also experiment with new ways to engage consumers on an ongoing basis. Zoovu, for example, tripled engagement rates by suggesting frequently asked questions to consumers, according to Taylor. An electronics retailer, for instance, may suggest that customers ask how many USB ports a laptop has.
“We find that a lot of customers don't even know what they're supposed to ask for. If you give them a blank sheet, then a lot of them will just get lost,” Taylor said. “Once they're engaged with the assistant … it will improve."
CX leaders should also be careful not to overplay the technology, as adoption remains low and forcing the experience on customers who prefer traditional search and filter risks alienating them. The assistant should be discoverable but shouldn’t crowd out the browsing experience that the majority of shoppers still prefer, experts said.
Looking ahead, 2026 will be “the year of the customer-led agentic experience” in retail, Schwartz said.
“We're going to see a lot of innovation happening in this space,” Schwartz said. “Whether it's agentic shopping inspiration, embedding carts or having a much more unified catalog across agentic channels.”
